Multiple reports describe a glitch in ChatGPT in which the chatbot became fixated on “goblins.” According to The Independent and The Wall Street Journal, the problem appears tied to efforts to prompt the system into adopting a particular “nerdy” or role-based personality. During testing, the model began producing unusually consistent goblin-themed language and word choices, even when the conversation did not call for that topic. The Wall Street Journal reports that OpenAI steps in to address the behavior after observing the pattern, indicating the issue is linked to how certain instructions or persona settings influence the model’s outputs. The Independent frames the incident as a potential concern because it highlights how seemingly harmless prompt tweaks or personality prompts can lead to repetitive or unexpected results. While the exact mechanism is not fully detailed in the available coverage, both outlets agree the behavior is abnormal, that it can be triggered by specific prompting, and that OpenAI takes corrective action to curb the problematic outputs. The episode underscores ongoing challenges in controlling and reliably constraining large language model behavior under different prompts.
OpenAI intervenes after ChatGPT repeatedly generates goblin-related responses
Multiple reports describe a glitch in ChatGPT in which the chatbot became fixated on “goblins.” According to The Independent and The Wall Street Journal, the problem appears tied to efforts to prompt...
- ChatGPT repeatedly generates goblin-related responses during certain prompts or persona settings.
- The issue is linked to attempts to steer the chatbot toward a “nerdy” or role-based personality.
- OpenAI intervenes after observing the goblin-fixation behavior.
- Multiple outlets describe the problem as abnormal and potentially problematic for reliable output control.
- The incident highlights how prompt or instruction changes can lead to unexpected repetition.
Strange bug with popular chatbot highlights underlying issues with AI
3 months agoThe company says attempts to make a “nerdy” personality for the bot led to odd word choices.
3 months ago
CXMT sues US Pentagon to remove blacklist designation tied to Chinese military support
ChangXin Memory Technologies (CXMT), a major Chinese DRAM chipmaker, has filed a lawsuit against the US Department of De...
Parents say raising teens is harder amid social media and constant oversight
Parents of teenagers describe feeling like they are “the police” as they try to manage screen time, online behaviour and...
Job seeker turns to AI training, earning up to $100 per hour
Mo Zohourian says he struggled to secure full-time employment and instead began working as an AI trainer after respondin...